Predicting gene ontology from a global meta-analysis of 1-color microarray experiments

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Global meta-analysis (GMA) of microarray data identifies genes with similar co-expression profiles.
• This method predicts gene function and phenotype, even for uncharacterized genes.
• GMA utilizes a guilt-by-association approach for predicting gene function.

Purpose of the Study:

• To define how sample size, dataset number, and ranking parameters impact prediction performance in GMA.
• To optimize GMA for accurate gene function prediction.

Main Methods:

• Downloaded 13,000 human 1-color microarrays from Gene Expression Omnibus (GEO) for GMA.
• Benchmarked prediction performance using Gene Ontology (GO) tree distance between predicted and annotated functions.
• Analyzed prediction performance across varying dataset sizes and gene set sizes.

Main Results:

• Prediction performance increases with more datasets, saturating around 2,000 experiments.
• Smaller gene sets yield higher precision but lower recall; larger sets increase recall and F-measure at the cost of precision.
• 72.5% of genes expressed in >=50 experiments received at least one predicted GO category.

Conclusions:

• GMA successfully predicted GO categories for 4,189 out of 5,720 unannotated genes.
• Approximately 17% of genes lacking GO predictions may have complex regulatory mechanisms.
• GMA is a powerful tool for accelerating the understanding of gene function and biological roles.